Experimental Physics and
Industrial Control System

Michael Davidsaver via Tech-talk <[email protected]> · Mon, 11 Mar 2019 20:31:23 -0700

Hi Casey,

On 3/8/19 4:18 PM, Layne (US), William C via Tech-talk wrote:
> am trying to narrow down which futex we are spending lots of time in

This got me thinking about how this would be done.  Looking at glibc
I think the first argument of the futex() syscall is the __data.__lock
member of pthread_mutex_t.

https://github.com/bminor/glibc/blob/43b1048ab9418e902aac8c834a7a9a88c501620a/nptl/pthread_mutex_lock.c#L43

If I change epicsMutexOsdShow() in posix/osdThread.c to:

> void epicsMutexOsdShow(struct epicsMutexOSD * pmutex, unsigned int level)
> {
> #if defined(__linux__) && defined(__GLIBC__)
>     /* GLIBC w/ NTPL is passing the &__lock as the first argument (UADDR)
>      * of the futex() syscall.
>      */
>     printf("    pthread uaddr=%p\n", &pmutex->lock.__data.__lock);
> #endif
> }

and apply

https://github.com/epics-base/epics-base/commit/d1149a0ba90096b3aa2890029ed9a430216d96ec

then I think the uaddr value shown below can be matched to futex() calls.
(I couldn't find a quick way to check this)

> epics> epicsMutexShowAll 0 1
> ellCount(&mutexList) 56 ellCount(&freeList) 0
> epicsMutexId 0x55db1e6f90e0 source ..//iocsh/iocsh.cpp line 78
>     pthread uaddr=0x55db1e6f90a0
> ...

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System